Failure mode: cannot convert float NaN to integer

[laldoroty]

sn="20000104"
ra="8.167274983711236"
dec="-43.08698868442529"
band="Y106"

After ~40 min.

Traceback (most recent call last):
  File "/global/cfs/cdirs/m4385/users/lauren/phrosty/phrosty/pipeline.py", line 559, in <module>
    main()
  File "/global/cfs/cdirs/m4385/users/lauren/phrosty/phrosty/pipeline.py", line 555, in main
    pipeline( args.through_step )
  File "/global/cfs/cdirs/m4385/users/lauren/phrosty/phrosty/pipeline.py", line 465, in __call__
    sfftifier = self.align_and_pre_convolve( templ_image, sci_image )
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/global/cfs/cdirs/m4385/users/lauren/phrosty/phrosty/pipeline.py", line 206, in align_and_pre_convolve
    sfftifier.resampling_image_mask_psf()
  File "/global/cfs/cdirs/m4385/users/lauren/sfft/sfft/SpaceSFFTCupyFlow.py", line 132, in resampling_image_mask_psf
    PixA_Eobj_GPU, EProjDict = CR.frame_extension(XX_proj_GPU=XX_proj_GPU,
                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/global/cfs/cdirs/m4385/users/lauren/sfft/sfft/utils/ResampKits.py", line 344, in frame_extension
    RMIN = (math.floor(XX_proj_GPU.min().item()) - 1) - KERHW[0]
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: cannot convert float NaN to integer

[laldoroty]

Have to verify this, but it seems like if it fails like this for one SN in one band, it fails like this for all the other bands as well.

[laldoroty]

This happens when the template and science image are the same image, which sometimes happens for the modified NN2 method that Bastien and I are working on.

[wmwv]

If it's an easy fix, sure, do it. But in general subtraction of two identical images is not a use case that most subtraction algorithms have as a test case. It's very easy to have divides by 0, lack of gradients in solutions, etc.